yacodebase-mcp 0.4.3

CLI + MCP server that indexes local codebases into Qdrant using OpenAI-compatible embeddings, then exposes semantic search to Claude Code and other MCP clients

yacodebase-mcp

Vector search MCP server for codebases. Index repos locally with AST-aware chunking; let Claude (or any MCP client) search them semantically and query their structure.

How it works

1. Index — walks repo files, chunks them using tree-sitter AST (function/method boundaries) with line-based fallback, embeds via OpenAI-compatible API, stores in in-process Qdrant.
2. Serve — exposes 12 MCP tools over stdio for search, structural analysis, and persistent knowledge.
3. Search — embeds the query, retrieves top-8 chunks across all indexed repos (or a specific one), returns ranked results with file path and line numbers.

Install

pip install yacodebase-mcp

Or from source:

uv tool install /path/to/yacodebase-mcp

For development:

uv sync

CLI

# Show version
yacodebase-mcp --version

# Index a repo (fails if already indexed)
yacodebase-mcp index ~/Code/myproject

# Re-index after changes (replaces existing index)
yacodebase-mcp reindex ~/Code/myproject

# Incrementally update index (only changed files)
yacodebase-mcp update ~/Code/myproject

# List indexed repos with chunk counts
yacodebase-mcp list

# Remove a repo from the index
yacodebase-mcp remove ~/Code/myproject

# Start MCP server (stdio, used by Claude Code)
yacodebase-mcp serve

Config commands

# Show current settings
yacodebase-mcp config list

# Set embedding model (known models auto-resolve vector size)
yacodebase-mcp config set embedding-model text-embedding-3-large
yacodebase-mcp config set embedding-model my-custom-model --vector-size 768

# Set API credentials
yacodebase-mcp config set api-key sk-...
yacodebase-mcp config set api-base https://my-provider.com/v1

# Revert a setting to default / env var fallback
yacodebase-mcp config unset embedding-model
yacodebase-mcp config unset api-key
yacodebase-mcp config unset api-base

Known models (vector size auto-detected):

Model	Vector size
text-embedding-3-small	1536
text-embedding-3-large	3072
text-embedding-ada-002	1536

Default: text-embedding-3-small.

Agent installation

Install the MCP server config into your dev agent's global settings with a single command. The installer resolves the absolute path to the binary so it loads regardless of the agent's PATH.

# Install into a specific agent
yacodebase-mcp install claude-code
yacodebase-mcp install cursor
yacodebase-mcp install windsurf
yacodebase-mcp install copilot        # GitHub Copilot (VS Code)
yacodebase-mcp install zed
yacodebase-mcp install opencode
yacodebase-mcp install codex          # OpenAI Codex CLI

# Install into all supported agents at once
yacodebase-mcp install all

# Preview changes without writing
yacodebase-mcp install claude-code --dry-run

# Check installation status
yacodebase-mcp install status

Config paths per agent and OS:

Agent	macOS	Linux	Windows
Claude Code	~/.claude.json	same	same
Cursor	~/.cursor/mcp.json	same	same
Windsurf	~/.codeium/windsurf/mcp_config.json	same	same
Copilot (VS Code)	~/Library/Application Support/Code/User/settings.json	~/.config/Code/User/settings.json	%APPDATA%\Code\User\settings.json
Zed	~/.config/zed/settings.json	same	%APPDATA%\Zed\settings.json
OpenCode	~/.config/opencode/config.json	same	%APPDATA%\opencode\config.json
Codex CLI	~/.codex/config.toml	same	same

Inject agent instructions (ensure always used)

Even with the MCP server installed globally, agents need explicit guidance to use search_codebase in a specific repo. The inject command writes instruction blocks into agent rule files inside the repo itself.

# Inject instructions for all agents into current repo
yacodebase-mcp inject run

# Inject for specific agents only
yacodebase-mcp inject run ~/Code/myproject --agent claude-code --agent cursor

# Preview without writing
yacodebase-mcp inject run --dry-run

# Remove injected instructions
yacodebase-mcp inject eject

# Check injection status
yacodebase-mcp inject status

Files written per agent:

Agent	File
Claude Code	CLAUDE.md
Cursor	.cursor/rules/codebase-search.mdc
GitHub Copilot	.github/copilot-instructions.md
Codex CLI	.codex/instructions.md

The injected block is idempotent (fenced with marker comments) and appends to existing content. eject removes only the injected block, leaving surrounding content intact.

Recommended workflow after indexing:

yacodebase-mcp index ~/Code/myproject
yacodebase-mcp inject run ~/Code/myproject

Shell completions

# bash — add to ~/.bashrc
source <(yacodebase-mcp completion bash)

# zsh — add to ~/.zshrc
source <(yacodebase-mcp completion zsh)

# fish — add to ~/.config/fish/config.fish
yacodebase-mcp completion fish | source

Manual Claude Code config

Add to ~/.claude.json at the top level (or use yacodebase-mcp install claude-code):

{
  "mcpServers": {
    "codebase-search": {
      "type": "stdio",
      "command": "/path/to/yacodebase-mcp",
      "args": ["serve"],
      "env": {}
    }
  }
}

API key can also be set via yacodebase-mcp config set api-key sk-... (persisted in ~/.yacodebase-mcp/settings.json), which takes precedence over the env var.

MCP tools

Search and navigation

search_codebase

Semantic search across indexed repos using embeddings.

Parameter	Type	Description
query	string	Natural language description of what to find
repo_path	string (optional)	Absolute path to a specific repo; omit to search all

Returns top-8 results ranked by similarity, each with file path, line range, score, and code block.

get_file_outline

Return the structural outline (functions, methods, classes) of a source file.

Parameter	Type	Description
file_path	string	Absolute path to the source file

search_symbols

Search for functions, methods, or classes by name across indexed repos.

Parameter	Type	Description
name	string	Symbol name or substring (case-insensitive)
repo_path	string (optional)	Absolute path to a specific repo; omit to search all

find_todos

Find TODO, FIXME, HACK, BUG, NOTE comments in indexed repos.

Parameter	Type	Description
repo_path	string (optional)	Absolute path to a specific repo; omit to search all

what_changed

Show files added or modified since the last index run.

Parameter	Type	Description
repo_path	string (optional)	Absolute path to a specific repo; omit to check all

list_indexed_repos

List all indexed repos with chunk count and last indexed timestamp. No parameters.

---

Knowledge persistence

Architectural decisions and notes persist in a local SQLite database across sessions.

add_decision

Record an architectural decision.

Parameter	Type	Description
title	string	Short title
body	string	Detailed explanation and rationale
category	string (optional)	e.g. architecture, security, performance

search_decisions

Search recorded decisions by keyword or category.

Parameter	Type	Description
query	string (optional)	Keyword to search in title and body
category	string (optional)	Filter by category

update_decision

Update the status of a decision.

Parameter	Type	Description
decision_id	int	ID from search_decisions output
status	string	active, superseded, implemented, or rejected

add_note

Save a note that persists across sessions.

Parameter	Type	Description
content	string	The note text
scope	string (optional)	project, file, or symbol
reference	string (optional)	File path or symbol name

get_notes

Retrieve saved notes.

Parameter	Type	Description
scope	string (optional)	Filter by scope; omit for all

---

Session orientation

session_bootstrap

Orient the agent at the start of a new session. Returns: indexed repos status, files changed since last index, active decisions, recent notes. Call this instead of reading files for orientation.

Parameter	Type	Description
repo_path	string (optional)	Scope to a specific repo; omit for all

---

Supported languages (AST chunking)

Language	Extensions	Chunk boundary
Python	.py	function_definition, decorated_definition
TypeScript	.ts	function_declaration, method_definition, arrow_function
TSX	.tsx	same as TypeScript
JavaScript	.js, .jsx	function_declaration, method_definition, arrow_function
Go	.go	function_declaration, method_declaration
Rust	.rs	function_item
Java	.java	method_declaration, constructor_declaration
HCL/Terraform	.tf	block

Files without AST support (.md, .yaml, .toml, .json, .rb, .cpp, .c, .h) fall back to 100-line sliding window with 20-line overlap.

Data storage

All data lives in ~/.yacodebase-mcp/:

~/.yacodebase-mcp/
  config.json      # indexed repo metadata (paths, repo_ids, chunk counts, file hashes, timestamps)
  settings.json    # embedding model, vector size, api_key, api_base
  knowledge.db     # SQLite: architectural decisions and notes
  qdrant/          # Qdrant in-process storage (one collection per repo)

Each repo gets a stable repo_id derived from its absolute path (used as Qdrant collection name). Reindexing replaces the collection in-place. Incremental updates (yacodebase-mcp update) use SHA256 hashes to skip unchanged files.

OpenAI-compatible providers

Set api-base to use any OpenAI-compatible embedding API (e.g. Ollama, vLLM, Azure):

yacodebase-mcp config set api-base http://localhost:11434/v1
yacodebase-mcp config set api-key ollama
yacodebase-mcp config set embedding-model nomic-embed-text --vector-size 768

After changing the model, reindex all repos (vector dimensions must match).